AI infrastructure faces copper and cooling crunch

The frenzy to build AI data center capacity is now exposing secondary supply chain constraints that threaten the pace of infrastructure deployment. Every megawatt of hyperscale data center infrastructure requires roughly 27 tonnes of copper for transformers, substations, power distribution, cooling systems, and high-capacity cabling. Yet China's export ban on sulphuric acid, a key input in copper refining, has disrupted refineries and spiked the cost of incremental supply. Silver, which shares similar refining pathways, has also surged; some traders report silver trading 26% above spot as supply dries up.

This has triggered secondary rotations. Copper itself has pushed to USD 13,619 per tonne on the LME, a fresh three-month high and only 6% below the January all-time peak. Traders are now hunting for copper-adjacent plays and junior mining names positioning for African acquisitions. Chinese state-owned Zhaojin Mining is scouting for gold and copper acquisitions in West Africa and Central Asia, signaling that tier-one nations are desperate to secure raw materials.

Beyond metals, the cooling and battery angle is emerging as critical infrastructure. SoftBank has invested billions in AI data center battery systems; energy storage plays like Eaton (BE), Flex Cells (FCEL), and zinc-halogen specialists Eos (EOSE) are attracting institutional attention. HVAC and energy integration plays like AAON are also being re-rated as the realization sets in that cooling is not a commodity afterthought but a mission-critical bottleneck. One broker highlighted that every major AI buildout now requires simultaneous investment in power grid expansion, cooling infrastructure, and energy storage.

The supply chain narrative is two-fold. On one hand, it validates the AI capex super-cycle thesis; constraints mean orders will persist and margins should hold for key suppliers. On the other hand, it introduces tail risk. If copper, cooling, or energy constraints become binding, hyperscalers may be forced to slow buildout or seek alternative architectures, potentially invalidating the most aggressive capex forecasts.